Micromachines 13 01252

micromachines
Article
Upright and Crawling Locomotion and Its Transition for a
Wheel-Legged Robot
Xuejian Qiu 1 , Zhangguo Yu 1,2 , Libo Meng 1,2, * , Xuechao Chen 1,2 , Lingxuan Zhao 1 , Gao Huang 3
and Fei Meng 1,2
1 School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China

2 Key Laboratory of Biomimetic Robots and Systems, Ministry of Education, Beijing 100081, China
3 Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
* Correspondence: menglibo@bit.edu.cn
Abstract: To face the challenge of adapting to complex terrains and environments, we develop a
novel wheel-legged robot that can switch motion modes to adapt to different environments. The
robot can perform efficient and stable upright balanced locomotion on flat roads and flexible crawling
in low and narrow passages. For passing through low and narrow passages, we propose a crawling
motion control strategy and methods for transitioning between locomotion modes of wheel-legged
robots. In practical applications, the smooth transition between the two motion modes is challenging.
By optimizing the gravity work of the body, the optimal trajectory of the center of mass (CoM) for
the transition from standing to crawling is obtained. By constructing and solving an optimization
problem regarding the posture and motion trajectories of the underactuated model, the robot achieves
a smooth transition from crawling to standing. In experiments, the wheel-legged robot successfully
transitioned between the crawling mode and the upright balanced moving mode and flexibly passed
a low and narrow passage. Consequently, the effectiveness of the control strategies and algorithms
proposed in this paper are verified by experiments.
Keywords: wheel-legged robot; two locomotion modes; mode transition; quadratic programming
Citation: Qiu, X.; Yu, Z.; Meng, L.;
Chen, X.; Zhao, L.; Huang, G.; Meng, F.
Upright and Crawling Locomotion
and Its Transition for a Wheel-Legged
Robot. Micromachines 2022, 13, 1252.
1. Introduction
https://doi.org/10.3390/mi13081252 As a masterpiece of species evolution, the physiological structure of the human body
provides a good structural basis for leg movement. A humanoid robot can walk on two legs
Academic Editor: Seokheun Choi
while freeing hands to complete complex two-arm collaborative tasks [1–7]. Human-sized
Received: 18 June 2022 robots are easier to adapt to the daily environment that is designed and decorated for
Accepted: 30 July 2022 humans. Considering the movement capabilities of the legs and arms and the adaptability
Published: 4 August 2022 to the human environment, we believe that a humanoid structure is the best morphological
Publisher’s Note: MDPI stays neutral
structure choice for robots. However, humanoid robots are less efficient at moving on
with regard to jurisdictional claims in flat roads than wheeled robots. Therefore, we decided to develop a novel wheel-legged
published maps and institutional affil- robot, which has the basic body structure and freedom of movement of a humanoid robot
iations. and replaces the feet with actuated wheels. Wheel-legged robots combine the advantages
of wheeled robots and legged robots [8–10], so they can travel efficiently on flat ground
with the help of wheels and overcome uneven terrains and obstacles with the help of
legs [11–16].
Copyright: © 2022 by the authors. In recent years, improving environmental adaptability has become the focus of research
Licensee MDPI, Basel, Switzerland. in the field of wheel-legged robots [17–21]. The wheel-legged robot is an underactuated
This article is an open access article system and its stability is greatly affected by the terrain and environment, so it is necessary
distributed under the terms and to have strong environmental adaptability. Handle [22], developed by Boston Dynamics,
conditions of the Creative Commons
is a milestone in the development of wheel-legged robots. Handle can pass through slopes,
Attribution (CC BY) license (https://
stairs, and jump over obstacles in upright balanced moving mode. Unfortunately, few
creativecommons.org/licenses/by/
technical details have been undisguised. Klemm et al. presented the wheel-legged robot
4.0/).
Micromachines 2022, 13, 1252. https://doi.org/10.3390/mi13081252 https://www.mdpi.com/journal/micromachines

Micromachines 2022, 13, 1252 2 of 21
Ascento that applied an LQR-assisted whole-body control strategy to stabilize the robot
system [23]. Ascento can jump up stairs by way of the mixed motion of legs and wheels.
H. Zhou et al. proposed a hydraulic wheel-legged robot WLR [24–26] that integrated leg
structures and hydraulic oil tubing to improve the movement ability and environmen-
tal adaptability of the robot. BIT developed a novel motor-driven wheel-legged robot
BHR-W [27] that can travel outdoors on grass and rough roads with the help of a layered
controller of legs and wheels. It is worth noting that the wheel-legged robots presented in
the above studies have only one mode of motion that allows the robot to perform upright
balanced moving in an open environment. However, in practical application scenarios, in
addition to the open environment, low and narrow passages are common and important
application environments, such as underground pipeline inspection and cave exploration.
Therefore, the environmental adaptability of the wheel-legged robot with only one motion
mode is limited.
In fact, if the robot can switch between different modes of locomotion according
to the different environments, the environmental adaptability of the wheel-legged robot
will be greatly improved, especially in unstructured environments coexisting with people.
In the DARPA robotics challenge, the Hubo robot [28,29] was equipped with wheels on
its knees and feet and it successfully completed the task with the help of the flexible
transition between walking mode and wheeled mode in complex environments. However,
its transition control strategy is full-drive and quasi-static, which is not suitable for an
underactuated wheel-legged robot. Liu et al. [30] developed a wheel-legged robot SR600-II
that has the structure of the foot and can perform flexible transitions between wheels and
feet. In another case, from the released video, the ANYmal robot can achieve the transition
from wheel-quadruped mode to wheel-biped mode by using the Multi-AMP algorithm
to enhance the deep learning framework [31,32]. Unfortunately, a long time is required to
generate motion priors with limited computing power, which is in most cases not available
for specific tasks. In general, the mode transition of wheel-legged robots mainly focuses
on the walking mode and the wheeled mode, while little research on the crawling mode
adapted to low and narrow environments has been published.
We believe that the ability of a wheel-legged robot to work in low and narrow envi-
ronments is an important indicator to measure its environmental adaptability. Therefore,
in this paper we develop a crawling motion mode of the wheel-legged robot for low and
narrow environments and propose control strategies for the transition between the crawling
mode and upright balanced moving mode. In particular, the standing recovery control of
the underactuated wheel-legged robot is a challenging technical task, that enables the robot
to transition from crawling to standing.
The main contributions of this paper are as follows:
• We develop a new wheel-legged robot with two locomotion modes and the ability to
transition between them and build simplified models for different motion modes of
the robot;
• We propose a crawling control strategy that can allow the robot to stably pass through
low and narrow passages with different heights and different curve radii;
• We propose locomotion mode transition control methods that enable the robot to
transition between the crawling mode and the upright balanced moving mode by
respectively optimizing the gravity work of the body and solving a quadratic pro-
gramming (QP) problem about the posture and motion trajectories of the robot.
The remainder of the paper is organized as follows: Section 2 introduces the mechani-
cal design and hardware of the robot system. Section 3 introduces the modeling and control
algorithms of the upright balanced moving mode and crawling mode of the robot. The
control methods of transition between locomotion modes are presented in Section 4. The
proposed approaches are verified by experiments on the BHR-W robot in Section 5. Finally,
we draw conclusions and discuss future work in Section 6.
2. BHR-W Robot System

2.1. Mechanical Design
The purpose of mechanical design is to develop a wheel-legged robot with human-like
proportions, range of movement and kinematic structure. Therefore, the kinematic scheme
for the robot is planned with 14 actuated DoFs based on human-like motion capabilities and
degrees of freedom configuration, as shown in Figure 1. Each leg has four DoFs including
hip pitch and roll, knee pitch and an actuated wheel. Each arm has three DoFs and a passive
wheel at the wrist. The robot is 1.61 m tall and the weight of the whole system is 55 kg.
400 mm
Figure 1. The wheel-legged robot BHR-W.
We analyse the range of movement of the robot’s arms and legs with the help of
kinematic analysis, as shown in Figure 2. The results show that the robot’s size and degrees
of freedom configuration enable the robot to have a large range of movement, which meet
the needs of human-like motion capabilities. The analysis results can assist researchers in
motion planning and control within a reasonable range of movement, as shown in Table 1.
Table 1. BHR-W joints angle range.
Joint Angle Range (deg.)

Shoulder pitch −90, +90
Shoulder roll 0, +80
Elbow pitch +30, +150
Hip pitch −90, +30
Hip roll −20, +60
Knee pitch 0, +135
－
－
－
－
－
－－
－－－
(a) (b)
Figure 2. Cloud maps of the range of movement of the robot’s limbs. (a) The cloud map of arms;
(b) The cloud map of legs.
The leg structure and freedom of movement configuration are designed in series as
shown in Figure 3a. To reduce the inertia of the leg and improve the response speed,
the leg motors are intensively placed in the hip and control the flexion, extension and
swing movements of the leg. We integrate the motor and reducer into the leg structure,
which increases the range of leg movement and reduces weight. The design scheme greatly
simplifies the leg structure and saves space that is originally prepared for the drive unit. The
torque that drives the motion of the knee joint comes from the motor torque transmitted by
the linkages. The length ratio of the torque transmission linkages is obtained by numerical
optimization, which can guarantee meeting the torque requirements of the knee joint in the
available range of movement.
(a) (b)
Figure 3. Three-dimensional (3D) structural model of the robot limbs. (a) Integrated structure of the
robot leg and motors; (b) Structure of the robot arm.
Each arm is divided into two parts: the upper arm and the forearm, as shown in
Figure 3b. The upper arm can perform pitch and roll movements. At the elbow joint, the
motor drives the ball screw mechanism to perform linear motion to drive the flexion and
extension of the forearm. The elbow joint can achieve a range of movement from 30 degrees
to 150 degrees. During the movement of the elbow joint, the angle δ between the screw
push rod and the forearm is constantly changing, which causes the torque acting on the
elbow joint is also changing. Therefore, we perform a kinematic analysis on the relationship
between elbow joint angle and torque at rated output torque condition, as shown in Figure 4.
It can be seen from the curve that in the range of movement from 50 degrees to 110 degrees,
the torque value of the elbow joint is larger than 45, which can guide the motion planning
of the elbow joint to make it move within a suitable torque range.
a
c
Figure 4. Kinematic analysis of the relationship between elbow joint angle and torque.
Actuated wheels at the end of each leg are directly driven by the hub motor, and
passive wheels are installed at the wrist. The actuated wheels provide power and stability
for the high-speed balanced moving, while the passive wheels assist the robot in changing
direction when crawling on all limbs. Additionally, the framework of the robot is mainly
fabricated from aluminium alloy, titanium alloy and carbon fiber to achieve high stiffness
and maintain lightweight.
2.2. Hardware System

The hardware system includes drive units, sensor units, communication systems and
a computer. Considering the mass of the robot and the impact from uneven ground, the
drive units need to provide a large torque. Therefore, we customize the frameless torque
motors and planetary reducers according to the size and torque requirements. All motors
communicate with the onboard computer via EtherCAT. The onboard computer is an Intel
NUC 8 and the processor is an i7-8559U. The inertial measurement unit(IMU) is installed
in the center of the torso and its feedback information plays an important role in the state
estimation of the robot. Furthermore, in order to help us evaluate the contact state between
the actuated wheels and the ground, a six-dimensional force sensor is integrated into
the ankle.
3. Modeling and Control of Two Locomotion Modes

The wheel-legged robot BHR-W has two motion modes: the upright balanced moving
mode on two wheels and the crawling mode on four limbs. Next, we construct the kinematic
and rigid body dynamic models of the above locomotion modes.
3.1. Coordinates
To clearly describe the locomotion of the robot, we define the coordinate systems and
variables as shown in Figure 5. The forward direction of the robot is the positive direction
of the x-axis. The vertical upward direction is the positive direction of the z-axis. The y-axis
is parallel to the wheel axis. θ denotes the tilt angle of the CoM relative to the z-axis. γ
denotes the angle between the forward direction of the robot and the y-axis at the inertial
coordinate system ∑0 .
θtorso
l
L1 x
CoM
γ
CoM
L2
z z
x
v z x
x
y
∑w y
L3 ∑0
(a) (b)
Figure 5. Generalized coordinates of the wheel-legged robot model: (a) The inverted pendulum
model on wheels; (b) The simplified model of body locomotion.
3.2. Modeling and Control of the Upright Balanced Moving Mode

3.2.1. Upright Balanced Moving Model
The upright balanced locomotion of the wheel-legged robot is a complex motion
formed by the coupling of wheel motion and body motion. Therefore, it is necessary to
build a layered control model for the robot [27].
The main purpose of wheel motion control is to achieve upright movement on the
basis of maintaining body balance and stability. A two-wheeled inverted pendulum
model [33–35] is built for the upright balanced moving control as shown in Figure 5a. The
torso, arms and legs are simplified as a rigid body on wheels. The height of the CoM and
the moment of inertia of the rigid body are provided by the body motion model.
Combining defined variables and the physical parameters of the model, the kinetic
and potential energies of the simplified model, T and V, can be derived.

T ( ϕ, ϕ̇) = Twheel ( ϕ̇) + Tbody ( ϕ, ϕ̇)
(1)
V ( ϕ) = Vwheel + Vbody ( ϕ),
where ϕ is the model generalized coordinate containing θ, φL and φR .

The Lagrangian energy function given by:
L( ϕ, ϕ̇) = T ( ϕ, ϕ̇) − V ( ϕ). (2)
The dynamic equations of the model are obtained by using the Euler–Lagrange equations:

d ∂L ∂L
− = M, (3)
dt ∂ ϕ̇ ∂ϕ
where t is the continuous time variable and M is the generalized torque.

Finally, the dynamic equation of the two-wheeled inverted pendulum model can be
expressed as:
Dw ( ϕ) ϕ̈ + Cw( ϕ,ϕ̇) ϕ̇ + Gw (ϕ) =

τw
(4)
τw = [ M f − ML − MR M f − ML M f − MR ] T ,
where ϕ = [θ φL φR ] T , Dw ∈ Rn×n denotes the mass matrix, Cw ∈ Rn denotes the vector

of coriolis and centrifugal terms, and Gw ∈ Rn is the vector of gravity term. MR and ML
respectively indicate the output torque of the two wheel motors. M f is the friction torque
on wheels.
Body motion can change the height of the CoM and the moment of inertia of the
two-wheeled inverted pendulum model. Therefore, we simplify the symmetrical body to a
3-link model as shown in Figure 5b, and the physical parameters are shown in Table 2.
Table 2. Parameters of the upright balanced moving model.
Symbol Parameter Name Value

Mb Mass of the body 51.5 kg
Mw Mass of the wheel 1.75 kg
Rw Radius of the wheel 127 mm
leg
L1 Length of the thigh 350 mm
leg
L2 Length of the calf 350 mm
l Length of the upper body 780 mm
L3 Distance between two wheels 328 mm
Iw Moment of inertia of wheel 0.0142 kg·m2
h Height of CoM -
hre f Reference height of CoM -
Iθ Moment of inertia of body around y-axis -
Iγ Moment of inertia of body around z-axis -
v Forward velocity -
vre f Reference forward velocity -
vturn Velocity compensation for turning -
φ̇L Rotation speed of left wheel -
φ̇R Rotation speed of right wheel -
The position of the CoM and the moment of inertia of the model can be expressed as:
i
∑i mi · PCoM ( β)
PCoM ( β) = (5)
M
I{θ,γ} = ∑ mi r{2iy,iz} , (6)

i
where PCoM ( β) is the position of the CoM, β = [ β 1 β 2 β 3 β 4 ] T is the generalized coordi-

nate of the 3-link model, and r{iy,iz} are the distance from the CoM of each part to the y-axis
or the z-axis in coordinate system ∑w .
3.2.2. Upright Balanced Moving Control

The wheel-legged robot is an underactuated system. A reliable balance controller
based on the linear quadratic regulator (LQR) and the body motion controller is developed
and applied to the upright balanced moving control of the wheel-legged robot, as shown in
Figure 6. Li et al. [36] proved that an LQR controller can provide highly reliable and robust
balance control for a two-wheeled inverted pendulum system.
The state equation of the robot system is obtained by linearizing the dynamic equations
of the simplified model in Equation (4).

ẋ = Ax + Bu
(7)
y = Cx,
where x = [θ φL φR θ̇ φ˙L φ˙R ] T , u = [ ML MR ] T , A ∈ R6×6 is the system matrix, B ∈ R6×2 is

the input matrix.
The problem solved by LQR controller is to obtain the optimal feedback gain matrix
K so that there is an optimal control variable u(t)∗ = −Kx (t) to minimize the quadratic
performance index function:
Z ∞
min J = x T (t) Qx (t) + u T (t) Ru(t) dt. (8)
0
By solving the discrete-time algebraic Riccati equation, the feedback gain matrix K of
the system is obtained:
A T P + PA − PBR−1 B T P + Q = 0 (9)
K = R−1 B T P, (10)
where Q and R are the weight matrices and P is the solution of the Riccati Equation (9).
Finally, the optimal input variable that can control the stability of upright balanced
locomotion is obtained:
u = −K x̂, (11)
where the optimal input variable u = [ ML MR ] T is the output torque of the two wheel
motors. The state variable of the system is x̂ supplied by the state estimator.
It is worth emphasizing that Q and R are constructed based on the importance of
the state variables and the controlled variables. The larger the value on the diagonal of
the Q matrix, the faster the system responds to the corresponding state variable. R is the
weight matrix of the controlled variables, and the larger the value on its diagonal is, the
stronger the constraint on the controlled variable. We gradually adjusted the weight value
in experiments until the robot system showed strong stability and robustness.
However, in the actual upright balanced movement on two wheels, the height of the
robot’s CoM h often changes due to body motions, such as squatting and standing up.
Therefore, we need to continuously adjust the parameters of the LQR balance controller to
ensure that the robot is stable and reliable in the upright balanced moving mode.
The controller calculates h the height of the CoM, Iγ the moment of inertia of body
around the z-axis and Iθ the moment of inertia of body around the wheel axis every five
periods with the help of the dynamics library.
q 2 2
h= x
PCoM ( β) z
+ PCoM ( β) (12)
y
Iθ = ICoM + Mh2 (13)
Iγ = z
ICoM x
+ M( PCoM ( β))2 , (14)
{ x,y,z}
where the value of ICoM are the moment of inertia of body around the CoM in { x, y, z}
direction given by the dynamics library.
+ Balance controller
－ LQR controller
u
Motion planning Upper body motion

State estimator controller controller
Crawling controller
+
－ Velocity controller
+
+ Wheel motors
Stand recovery Kneeling down Turning controller
controller controller +
variable height +
Mode transition controller Joint motors
controller
Figure 6. Overview of the controller architecture including two-wheeled balance controller, crawling
on four limbs controller and mode transition controller. α = [α1 α2 α3 ] T is the angle of each joint of
the arm.
They are fed into the LQR balance controller to obtain the real-time feedback gain ma-
trix K̃. Finally, the balance controller uses the K̃ matrix and x̂ fed back by the state observer
to obtain the real-time optimal input variable ũ for the upright balanced moving mode.
ũ = −K̃ x̂. (15)
3.3. Modeling and Control of the Crawling Mode

3.3.1. Crawling Motion Model
The main goal of crawling on four limbs is to pass through low and narrow passages.
It is necessary for the crawling motion model to have the ability to change the height of the
torso from the ground and to achieve flexible turning. The crawling motion model with the
help of the hybrid locomotion of four limbs is shown in Figure 7. The dimensions of each
body part are shown in Table 3.
The height of the torso from the ground Htorso can be derived as:
leg leg
Htorso = L1arm sin(α2 ) + L2arm sin(α2 + α3 ) + L aw = L1 sin( β 2 ) + L2 sin( β 2 + β 3 ) + Rw . (16)
The turning radius Rturn can be expressed as:
LT
Rturn = (17)
sin β 1
L T = Ltorso + L arm + Lleg cos β 1 (18)
L arm = − L1arm cos α2 − L2arm cos(α2 + α3 ) (19)
leg leg
Lleg = L1 cos β 2 + L2 cos( β 2 + β 3 ). (20)
Table 3. Parameters of the crawling motion model.
Symbol Parameter Name Value

L1arm Length of the upper arm 400 mm
L2arm Length of the forearm 320 mm
leg
L1 Length of the thigh 350 mm
leg
L2 Length of the calf 350 mm
Ltorso Length of the torso 500 mm
Haw Height from ground to wrist 110 mm
Htorso Height of the torso -
Rturn Radius of turning -
Stable
Region
(a) (b)
Figure 7. The model of limbs crawling motion: (a) The variable height crawling model; (b) The
crawling turning model.
3.3.2. Crawling Motion Control

When crawling as shown in Figure 7, wheels equipped at the ends of four limbs of
the robot touch the ground. The state of touching the ground at four points is reliable and
stable. Therefore, in this mode, we do not perform additional control for stability and only
control the movement of the robot such as forward, backward, turning, etc.
In the crawling motion mode, the drive wheel motors are in the speed control mode.
When driving on a straight road, the controller will drive the wheel motors to reach the
reference velocity v. When passing through a curve, the operator can control and drive the
roll angle of the hip joint to make the robot turn. During turning as shown in Figure 7b,
the angle between the upper body and the legs is less than 180 degrees and the CoM is
in the stable region. If the radius of the curve is less than the minimum turning radius
of the robot, the differential driving of the wheels can be activated to assist the robot in
successfully passing through the curve successfully.
In addition, to enable the robot to pass through passages with different heights, the
distance Htorso between the torso and the ground can also be controlled by changing the
each joint angle of the four limbs. The passive wheels are installed at the wrist, so it
is necessary to ensure that the forearm and calf are parallel to the ground as shown in
Figure 7a:
α3 = π − α2
(21)
β3 = − β2 .
By combining Equations (16) and (21), we can obtain each joint angle of the four limbs,
which can help the robot to pass through passages of different heights.
4. Modeling and Control of Mode Transition

4.1. Modeling and Control of Kneeling
4.1.1. Kneeling Model
The kneeling process is a transition from the upright balanced moving mode to the
kneeling state, as shown in Figure 8. The kneeling model of the robot is the same as
Figure 5b, in which the upper body and arms are simplified as a rigid body. After kneeling,
the angle between the calf and the ground is determined:
Rw
β 4 = arcsin( leg
). (22)
L2
The relative position of the CoM of the simplified model can be expressed as:
{ix,iz}
{ x,z} ∑i mi · PCoM ( β)
PCoM ( β) = , (23)
M
where β is the joint angle of the kneeling down model, and M is the total mass of the robot.
{ x,z} i
PCoM (q) is the position of the CoM after kneeling down, PCoM (q) represent the position of
the CoM of each part, and mi represent the mass of each part.
x
Stable Region
Figure 8. The model of kneeling down.
4.1.2. Kneeling Control

The kneeling process of the underactuated wheel-legged robot is similar to a forward
falling process as shown in Figure 8. During this process, the controller stops the wheels
and causes the robot to lean forward and fall. Therefore, in order to maintain the stability
of the system, the mode transition controller needs to ensure that the CoM of the robot is
located in the stable region formed by the wheels and the knee after kneeling.
x x x
pmin ≤ PCoM ( β) ≤ pmax , (24)
where pminx x
and pmax are the boundary coordinates of the stable region.
It is necessary to ensure that the velocity of the CoM in the x-axis direction is as small
as possible to prevent further falling. In addition, we need to optimize the velocity of the
CoM in the z-axis direction to reduce the impact force on the knee joint.
z x

minwz ṖCoM ( β̇, β) + wx ṖCoM ( β̇, β) (25)
z z x x

⇒ minwz ( P̂CoM ( β̂) − P CoM( β)) + wx ( P̂ ( β̂) − P
CoM ( β)),
CoM (26)
where wz and wx are the weights of the velocity in the z-axis direction and the x-axis
{ x,z}
direction respectively, and P̂CoM ( β̂) is the initial position of the CoM.
We construct an optimization problem about the average velocity of the CoM during
the kneeling process, and we obtain the optimal kneeling position and posture that satisfies
various demand constraints:
z z x x

min wz ( P̂CoM ( β̂) − PCoM ( β)) + wx ( P̂CoM ( β̂) − PCoM ( β))
 x x x
 pmin ≤ PCoM ( β) ≤ pmax

(27)
s.t. β min ≤ β ≤ β max
 β 4 = arcsin( Rleg
 w
).
L2
4.2. Modeling and Control of Standing Recovery

4.2.1. Standing Recovery Model
The standing recovery process is the transition from the crawling mode to the upright
balanced moving mode. The most critical motion in this process is from kneeling to standing
as shown in Figure 9. The entire kneeling-to-standing motion is similar to swinging up and
stabilizing a second-order inverted pendulum. In this process, there is no relative motion
between drive wheels and the calf. Wheels only roll on the ground. Therefore, we simplify
the robot model to a special second-order inverted pendulum with an underactuated joint
that hinges on the ground. The thigh and calf structures are regarded as the first-order link,
and the torso and arms are regarded as the second-order link.
Center of Mass
Second Link
Drive Motor
First Link
Underactuated Joint
(a) (b) (c)
Figure 9. An under-actuated second-order inverted pendulum model: (a) The kneeling posture;
(b) The leaning back posture; (c) The squatting posture.
It is assumed that the left and right sides of the robot system are completely symmetri-
cal and the mass distribution is uniform. Additionally, using the Euler–Lagrange equations,
the dynamic equation of the special inverted pendulum can be derived as:
Dr (q)q̈ + Cr (q, q̇)q̇ + Gr (q) = Sr τr , (28)
where q = [q1 q2 ] T is the joint angle of the under-actuated link model, τr is the torque of
all joints, Sr ∈ Rn×n is the control selection matrix.
4.2.2. Standing Recovery Control

The motion planning and control of standing recovery is inspired by the instinctive
motions of humans from crawling to standing up. First, a predefined and planned action
sequence is triggered from the motion library. During the execution of the action, the
trajectory is locally online optimized according to the starting conditions of the balance
controller. Therefore, the standing recovery controller is composed of a feed-forward
motion library and a trajectory optimization controller. In the following, the standing
recovery controller of the robot will be described in further detail.
From Crawling to Kneeling: Inspired by human actions and considering the DoFs of
the robot, the humanoid limb motion trajectories are planned and stored in the motion
library. The controller makes the joint motors follow the preplanned specific trajectories, so
that the robot performs the mode transition to the kneeling state as shown in Figure 10.
(a) (b) (c) (d) (e)

Figure 10. The transition process from crawling to kneeling (a–e).
From Kneeling to Standing: Based on the simplified underactuated inverted pendulum

model and humanoid motion, we divide the transition from kneeling to standing into
two stages.
The purpose of the first-stage control, as shown in Figure 9a,b, is to move the CoM
of the robot back to above the wheel axis vertically and to make the knee joint leave the
ground. Therefore, we need to plan the motions and control the upper body of the robot to
lean back according to the desired position of the CoM. Then, we use the kinematic method
to map the CoM trajectory to all joints.
b b
ṖCoM = JCoM q̇, (29)
where b ṖCoM is the velocity of the CoM in the coordinate system ∑w , b JCoM is the Jacobian
matrix of the CoM, and q̇ is the joint angular velocity.
The upper body of the robot performs a backward leaning motion to move the CoM
of the robot backward. Due to the existence of underactuated joint, when the CoM reaches
above the wheel axis, the robot will lean back around the underactuated joint with a certain
angular velocity.
Since the robot still has a certain angular velocity rotating around the underactuated
joint, the system state is unstable, so the second-stage control takes effect. The purpose
of the second-stage control is to adjust the robot posture to meet the starting conditions
of the balance controller. The mechanical median position is a special posture where the
robot can self-balance without control and the CoM is vertically above the wheel axis. The
two-wheeled balance controller needs to be triggered when the robot posture is close to
the mechanical median position, otherwise the robot may oscillate due to overshoot. In
addition, if the robot still has a large angular velocity rotating around the wheel axis when
the balance controller is activated, it will cause a huge challenge to the stability of the robot.
Therefore, the goal of the standing recovery controller is to ensure that robot reaches the
mechanical median position and has no angular velocity around the wheel axis. Although
the CoM of the robot is above the wheel axis vertically at the end of the first-stage control,
the robot has a relatively large angular velocity rotating around the wheel axis. Clearly,
this does not satisfy the starting conditions of the balance controller. Therefore, the state
estimator needs to feed back the angular velocity information of the robot rotating around
the wheel axis as the state parameters for the second-stage control.
Because the robot has angular velocity of backward rotation, the CoM will move back
away from the mechanical median position. Each control cycle predicts whether the CoM
of the robot will deviate from the mechanical median position in the next cycle based on the
position of the CoM and the angular velocity information. If it is predicted that there will
be a large deviation between the position of the CoM and the mechanical median position,
the controller will plan the trajectory online to make the CoM return back. If the deviation
is small enough to be within a certain threshold range, it is considered that the starting
condition of the balance controller is satisfied.
Predict the angle of the underactuated joint in the next cycle and plan the motion
trajectory of the actuated joint based on it to make the robot reach back to the mechanical
median position. Since the model is underactuated, the robot cannot accurately perform
the planned trajectory, so we need to use the optimization method to obtain the optimal
trajectory and torque of the actuated joints combined with the online planning trajectory
and the dynamic Equation (28).
By constructing and solving a QP problem [37] about the posture and motion trajecto-
ries of the underactuated model, we obtain the optimal joint trajectory and torque for the
robot to transition from crawling to standing.
1 1 1
min k AX − bk2 = X T ( A T A) X + (− A T b)T X + b T b
2 2 2

[ C ( q, q̇ ) D ( q ) − S 1 ] X + G ( q ) = 0
(30)


q̇min ≤ S2 X ≤ q̇max

s.t.
τ ≤ S3 X ≤ τmax
 min


qmin ≤ q ≤ qmax .
The optimization variable is X = [q̇ q̈ τ ] T . Si is the selection matrix and the equation
constraint is the dynamic equation of the underactuated model, which ensures that the
motion of the underactuated joint follows the laws of physics. Inequality constraints
provide the range of joint angles and torque limits.
In practical applications, we use Eigen-QuadProg [38] to solve the QP problem online.
When the predicted deviation of the centroid position from the mechanical median position
is small enough to be within a certain threshold range, the robot will end the trajectory
optimization control and start the balance controller.
The purpose of the cost function is to ensure that the motion of the robot is close to the
desired position and velocity, as follows:
A = diag(wq̇1 , wq̇2 , wq̈1 , wq̈2 , wτ1 , wτ2 )

   
O2×1 O2×1
 re f   re f re f  (31)
 q̈   k (q − q1 ) + k d1 (q̇1 − q̇1 ) 
b =  1re f  =  p1 1
 q̈  k p2 (q2 − qre f ) + k d2 (q̇2 − kq̇1 ),

2 2
O2×1 O2×1
where wi represents the weight of the specific desired behavior. The penalty for deviating
from the desired state is reflected in the acceleration term. qre f and q̇re f from motion
trajectories are planned by a higher level controller. In addition, k represents the specific
angular velocity relationship between the drive rod and the under-actuated rod, whose
purpose is to restrain the angular velocity of the system as small as possible.
When the robot reaches the mechanical median position, the CoM is vertically above
the wheel axis. Therefore, at this moment, the angular momentum of the under-actuated
system is conserved.
∑ Ii ωi = Iω, (32)
i
where Ii is the moment of inertia of each part around the under-actuated joint, ωi is the
angular velocity of each part, I is the moment of inertia of the whole system around the
under-actuated joint, and ω is the angular velocity of the whole system.
ω as small as possible is one of the starting conditions of the balance controller. So the
angular velocity relationship k can be expressed as:
I1
k= . (33)
I2
5. Experiments
In this section, we introduce the traveling experiments of BHR-W in two locomotion
modes and the mode transition experiments between them. The control rate of the robot is
1 kHz. The motion commands for the robot, such as forward, backward and turn, and the
mode transition commands are all controlled by the operator remotely.
5.1. Upright Balanced Moving Experiment

We control the BHR-W to carry out the upright balanced moving experiment on an
outdoor road. Figure 11 presents snapshots of the upright balanced moving experiment.
The robot successfully completes the motion commands input by the operator and main-
tains the balance and stability of the robot system during the moving process with the help
of the balance controller. Motion commands are input remotely by the operator, and the
balance controller runs autonomously on the onboard computer.
(a) (b) (c) (d) (e) (f)

Figure 11. Upright balanced moving experiment on the outdoor road: (a–c) Forward experiment;
(d–f) Turning experiment.
The data curves are shown in Figure 12, which shows that the operator controls the
robot to accelerate forward and then decelerate, and turn in the balanced standing state.
The value of θ shows that the body leans forward when accelerating and leans backward
when decelerating. The experiment shows that the balance controller is stable and reliable.
－－
－－
Figure 12. Date graph of the upright balanced moving experiment on the outdoor road: v is the
forward velocity of the robot; θ is the angle between the CoM and the z-axis.
5.2. Crawling Experiment

To demonstrate that the robot can pass through low and narrow passages and to show
the reliability of the proposed motion control strategy for crawling, we perform a crawling
experiment. As shown in Figure 13, with the assistance of the operator, the BHR-W passes
through a channel with a size of 30 cm × 80 cm × 60 cm and makes a large radius turn.
(a) (b) (c) (d)
(e) (f) (g) (h)

Figure 13. The robot crosses a channel with a size of 30 cm × 80 cm × 60 cm and performs a turn (a–h).
5.3. Kneeling Experiment

The robot completes the transition from the upright balanced moving mode to the
kneeling posture as shown in Figure 14. In the experiment, we find that the tilting and
falling direction of the robot can not be accurately controlled after the wheel motor stops.
Therefore, we control the wheel motor to rotate backward by a small angle, so that the
robot has a tendency to fall forward. Figure 15 shows that the position of the CoM is in
x
the stable region (PCoM > 0.078 m) after kneeling on the ground. At the moment of hitting
x
the ground, v x and vz are within the acceptable range. The oscillation of PCoM z
and PCoM
is caused by the elastic deformation of the tyre. At the beginning of the experiment, the
sudden change in v x is caused by the backward rotation of the wheel motor. In summary,
the controller stably completes the mode transition task from the upright balanced moving
mode to the kneeling state and reduces the impact force on the knee joint.
(a) (b) (c) (d) (e)

Figure 14. The transition experiment of the robot from the upright balanced moving mode to kneeling
on the ground: (a) Balanced standing; (b–e) From balanced standing to kneeling.
(a)
Stable Region
(b)
－
(c)
(d)
Figure 15. The position and velocity data of the CoM during the kneeling down experiment (a–d).
5.4. Standing Recovery Experiment

First, the prototype performs the preplanned motion trajectories to complete the
transition from crawling on four limbs to kneeling as shown in Figure 16.
(a) (b) (c) (d) (e)

Figure 16. The transition of robot posture from crawling on all limbs to kneeling: (a) Crawling on
limbs; (b–e) From crawling to kneeling.
Then, with the help of the mode transition controller, the BHR-W transitions from
kneeling to the upright balanced moving mode as shown in Figure 17. During the process
shown from Figure 17a–d, the controller causes the upper body to lean back and leaves the
knee joint off the ground. During the process shown from Figure 17e,f, with the assistance
of the QP online controller, the posture of the underactuated robot reaches the mechanical
median position and the balance controller is activated.
(a) (b) (c) (d) (e) (f)
Figure 17. Mode transition experiment from kneeling to balanced standing (a–f ).
As seen in Figure 18, the CoM moves rapidly backwards and eventually stabilizes at
approximately 0. PCoMx = 0 means that the CoM is vertically above the wheel axis, which
proves that the robot posture reaches the mechanical median position. The velocity curve
shows that v x is close to 0 when arriving at the mechanical median position. The above
data analysis shows that the state of the robot satisfies the starting conditions of the balance
controller, so the position of the CoM does not have a large oscillation. In general, the
experiment demonstrates the effectiveness of the standing recovery controller.
(a)
(b)
Figure 18. The position and velocity data of the CoM during the modal transition from kneeling to
x
balanced standing (a,b). PCoM is the projection of the distance of the CoM relative to the coordinate
system ∑w on the x-axis and v x is the velocity of the CoM in the x-axis direction.
Finally, the experiment of the robot standing up, which changes the height of the CoM,
is shown in Figure 19. Figure 20 records the position and velocity data of the CoM relative
to the coordinate system ∑w during the standing process. The height of the CoM steadily
increases from 0.48 m to 0.72 m. PCoM x fluctuates around 0 mm and finally stabilizes at
approximately 2 mm, and v x does not exceed ±0.03 m/s. In summary, the data show that
the CoM of the robot has no obvious overshoot during standing up.
The balance controller combines the upper body motion controller and the LQR
controller to update the feedback gain matrix K every control cycle to achieve stable balance
control in response to the robot’s centroid height variation in standing up experiment, which
proves that the balance controller has a certain robustness.
(a) (b) (c) (d) (e)

Figure 19. The experiment of the robot standing up (a–e).
(a)
－
(b)
(c)
Figure 20. The position and velocity data of the CoM relative to the coordinate system ∑w during
standing up (a–c).
6. Conclusions
This paper presents a novel wheel-legged robot BHR-W and control algorithms for
two locomotion modes and mode transitions. Through the cooperation of the LQR balance
controller and the body motion controller, BHR-W can achieve stable and efficient upright
balanced locomotion on two wheels. The crawling control algorithm allows the robot
to flexibly pass through low and narrow passages with different heights and different
curve radii. In addition, the mode transition of the robot can be realized by the optimal
control of the underactuated model. In the experiments, the robot completes the locomotion
of the upright balanced moving mode and the crawling mode and the mode transition
between them. The effectiveness and robustness of the proposed control algorithms have
been validated by the experimental results. We believe that it is necessary for the robot
to switch motion modes when facing different terrains and environments. Therefore, the
control methods of the wheel-legged robot proposed in this paper have strong potential for
practical application.
In the future, we will equip the robot with a vision system so that it can autonomously
recognize different environments and switch motion modes to adapt to the environment.
Author Contributions: Conceptualization, X.Q. and Z.Y.; methodology, X.Q. and L.Z.; hardware, G.H.
and X.Q.; validation, X.Q. and L.Z.; investigation, L.M. and F.M.; writing—original draft preparation,
X.Q.; writing—review and editing, X.Q., Z.Y. and L.M.; project administration, Z.Y. and X.C. All
authors have read and agreed to the published version of the manuscript.
Funding: This research was funded in part by the National Key Research and Development Project
under Grant No. 2018YFE0126200, in part by the National Natural Science Foundation of China
under Grant No. 62073041 and No. 62103053, and in part by the “111” Project under Grant B08043.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
Abbreviations
The following abbreviations are used in this manuscript:
LQR Linear Quadratic Regulator

DoF Degrees of Freedom
QP Quadratic Programming
CoM Center of Mass
BIT Beijing Institute of Technology
EtherCAT Ether Control Automation Technology
IMU Inertial Measurement Unit
References
1. Atlas. Available online: https://www.bostondynamics.com/atlas (accessed on 6 June 2022).
2. Henze, B.; Roa, M.A.; Ott, C. Passivity-Based Whole-Body Balancing for Torque-Controlled Humanoid Robots in Multi-Contact
Scenarios. Int. J. Robot. Res. 2016, 35, 1522–1543. [CrossRef]
3. Henze, B.; Dietr, A.; Ott, C. An approach to combine balancing with hierarchical whole-body control for legged humanoid robots.
IEEE Robot. Autom. Lett. 2016, 1, 700–707. [CrossRef]
4. Kanoulas, D.; Lee, J.; Caldwell, D.G.; Tsagarakis, N.G. Center-of-mass-based grasp pose adaptation using 3d range and
force/torque sensing. Int. J. Hum. Robot. 2018, 15, 1850013. [CrossRef]
5. Gori, I.; Pattacini, U.; Tikhanoff, V.; Metta, G. Ranking the good points: A comprehensive method for humanoid robots to grasp
unknown objects. In Proceedings of the 2013 IEEE International Conference on Advanced Robotics (ICAR), Montevideo, Uruguay,
25–29 November 2013; pp. 1–7. [CrossRef]
6. Wang, Y.; Li, W.; Togo, S.; Yokoi, H.; Jiang, Y. Survey on Main Drive Methods Used in Humanoid Robotic Upper Limbs. Cyborg
Bionic Syst. 2021, 2021, 9817487. [CrossRef]
7. Namiki, A.; Yokosawa, S. Origami folding by multifingered hands with motion primitives. Cyborg Bionic Syst. 2021, 2021, 9851834.
[CrossRef]
8. Gong, Y.; Hartley, R.; Da, X.; Hereid, A.; Harib, O.; Huang, J.-K.; Grizzle, J. Feedback control of a cassie bipedal robot: Walking,
standing, and riding a segway. In Proceedings of the 2019 American Control Conference (ACC), Philadelphia, PA, USA,
10–12 July 2019; pp. 4559–4566. [CrossRef]
9. Li, Q.; Yu, Z.; Chen, X.; Zhou, Q.; Zhang, W.; Meng, L.; Huang, Q. Contact force/torque control based on viscoelastic model for
stable bipedal walking on indefinite uneven terrain. IEEE Trans. Autom. Sci. Eng. 2019, 16, 1627–1639. [CrossRef]
10. Dong, C.; Yu, Z.; Chen, X.; Chen, H.; Huang, Y.; Huang, Q. Adaptability control towards complex ground based on fuzzy logic for
humanoid robots. IEEE Trans. Fuzzy Syst. 2022, 30, 1574–1584. [CrossRef]
11. Klemm, V.; Morra, A.; Salzmann, C.; Tschopp, F.; Bodie, K.; Gulich, L.; Küng, N.; Mannhart, D.; Pfister, C.; Vierneisel, M.
Ascento: A two-wheeled jumping robot. In Proceedings of the 2019 International Conference on Robotics and Automation
(ICRA), Montreal Convention Center, Montreal, QC, Canada, 20–24 May 2019; pp. 7515–7521. [CrossRef]
12. Bjelonic, M.; Sankar, P.K.; Bellicoso, C.D.; Vallery, H.; Hutter, M. Rolling in the deep–hybrid locomotion for wheeled-legged robots
using online trajectory optimization. IEEE Robot. Autom. Lett. 2020, 5, 3626–3633. [CrossRef]
13. Klamt, T.; Behnke, S. Anytime hybrid driving-stepping locomotion planning. In Proceedings of the 2017 IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS), Vancouver, BC, Canada, 24–28 September 2017; pp. 4444–4451. [CrossRef]
14. Bjelonic, M.; Grandia, R.; Harley, O.; Galliard, C.; Zimmermann, S.; Hutter, M. Whole-body mpc and online gait sequence
generation for wheeled-legged robots. In Proceedings of the 2021 IEEE/RSJ International Conference on Intelligent Robots and
Systems (IROS), Prague, Czech Republic, 27 September–1 October 2021; pp. 8388–8395. [CrossRef]
15. Wang, L.; Meng, L.; Kang, R.; Liu, B.; Gu, S.; Zhang, Z.; Meng, F.; Ming, A. Design and Dynamic Locomotion Control of
Quadruped Robot with Perception-Less Terrain Adaptation. Cyborg Bionic Syst. 2022, 2022, 9816495. [CrossRef]
16. Li, J.; Wang, J.; Peng, H.; Hu, Y.; Su, H. Fuzzy-torque approximation-enhanced sliding mode control for lateral stability of mobile
robot. IEEE Trans. Syst. Man Cybern. Syst. 2021, 52, 2491–2500. [CrossRef]
17. Kashiri, N.; Baccelliere, L.; Muratore, L.; Laurenzi, A.; Ren, Z.; Hoffman, E.M.; Kamedula, M.; Rigano, G.F.; Malzahn, J.;
Cordasco, S.; et al. Centauro: A hybrid locomotion and high power resilient manipulation platform. IEEE Robot. Autom. Lett.
2019, 4, 1595–1602. [CrossRef]
18. Wang, S.; Cui, L.; Zhang, J.; Lai, J.; Zhang, D.; Chen, K.; Zheng, Y.; Zhang, Z.; Jiang, Z.-P. Balance control of a novel wheel-legged
robot: Design and experiments. In Proceedings of the 2021 IEEE International Conference on Robotics and Automation (ICRA),
Sanya, China, 6–9 December 2021; pp. 6782–6788. [CrossRef]
19. Cui, L.; Wang, S.; Zhang, J.; Zhang, D.; Lai, J.; Zheng, Y.; Zhang, Z.; Jiang, Z.-P. Learning-based balance control of wheel-legged
robots. IEEE Robot. Autom. Lett. 2021, 6, 7667–7674. [CrossRef]
20. Chen, H.; Wang, B.; Hong, Z.; Shen, C.; Wensing, P.M.; Zhang, W. Underactuated motion planning and control for jumping with
wheeled-bipedal robots. IEEE Robot. Autom. Lett. 2020, 6, 747–754. [CrossRef]
21. Xin, S.; Vijayakumar, S. Online dynamic motion planning and control for wheeled biped robots. In Proceedings of the 2020
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Las Vegas, NV, USA, 24–30 October 2020;
pp. 3892–3899. [CrossRef]
22. Handle. Available online: https://www.bostondynamics.com/handle (accessed on 6 June 2022).
23. Klemm, V.; Morra, A.; Gulich, L.; Mannhart, D.; Rohr, D.; Kamel, M.; de Viragh, Y.; Siegwart, R. LQR-Assisted Whole-Body
Control of a Wheeled Bipedal Robot With Kinematic Loops. IEEE Robot. Autom. Lett. 2020, 5, 3745–3752. [CrossRef]
24. Zhou, H.; Li, X.; Feng, H.; Li, J.; Zhang, S.; Fu, Y. Model decoupling and control of the wheeled humanoid robot moving in sagittal
plane. In Proceedings of the 2019 IEEE-RAS 19th International Conference on Humanoid Robots (Humanoids), Toronto, ON,
Canada, 15–17 October 2019; pp. 1–6. [CrossRef]
25. Li, X.; Zhang, S.; Zhou, H.; Feng, H.; Fu, Y. Locomotion adaption for hydraulic humanoid wheel-legged robots over rough
terrains. Int. J. Hum. Robot. 2021, 18, 2150001. [CrossRef]
26. Li, X.; Zhou, H.; Feng, H.; Zhang, S.; Fu, Y. Design and experiments of a novel hydraulic wheel-legged robot (WLR). In Proceedings
of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Madrid, Spain, 1–5 October 2018;
pp. 3292–3297. [CrossRef]
27. Zhao, L.; Yu, Z.; Chen, X.; Huang, G.; Wang, W.; Han, L.; Qiu, X.; Zhang, X.; Huang, Q. System design and balance control of a
novel electrically-driven wheel-legged humanoid robot. In Proceedings of the 2021 IEEE International Conference on Unmanned
Systems (ICUS), Beijing, China, 1–5 October 2021; pp. 742–747. [CrossRef]
28. Oh, P.; Sohn, K.; Jang, G.; Jun, Y.; Cho, B.-K. Technical overview of team drc-hubo@ unlv’s approach to the 2015 darpa robotics
challenge finals. J. Field Robot. 2017, 34, 874–896. [CrossRef]
29. Hobo: [DRC 2015] Team KAIST Full Video. Available online: https://www.youtube.com/watch?v=PomkJ4l9CMU (accessed on
7 June 2022).
30. Liu, T.; Zhang, C.; Wang, J.; Song, S.; Meng, M.Q.-H. Towards terrain adaptablity: In situ transformation of wheel-biped robots.
IEEE Robot. Autom. Lett. 2022, 7, 3819–3826. [CrossRef]
31. ANYmal. Available online: https://www.youtube.com/watch?v=kEdr0ARq48A (accessed on 7 June 2022).
32. Vollenweider, E.; Bjelonic, M.; Klemm, V.; Rudin, N.; Lee, J.; Hutter, M. Advanced Skills through Multiple Adversarial Motion
Priors in Reinforcement Learning. In Proceedings of the 2022 IEEE/RSJ International Conference on Intelligent Robots and
Systems (IROS), Kyoto, Japan, 23 October 2022. [CrossRef]
33. Pathak, K.; Franch, J.; Agrawal, S.K. Velocity and position control of a wheeled inverted pendulum by partial feedback lineariza-
tion. IEEE Trans. Robot. 2005, 21, 505–513. [CrossRef]
34. Huang, J.; Guan, Z.-H.; Matsuno, T.; Fukuda, T.; Sekiyama, K. Sliding-mode velocity control of mobile-wheeled inverted-
pendulum systems. IEEE Trans. Robot. 2010, 26, 750–758. [CrossRef]
35. Zhou, Y.; Wang, Z.; Chung, K.-W. urning motion control design of a two-wheeled inverted pendulum using curvature tracking
and optimal control theory. J. Optim. Theory Appl. 2019, 181, 634–652. [CrossRef]
36. Li, Z.; Yang, C.; Fan, L. Advanced Control of Wheeled Inverted Pendulum Systems; Springer Science & Business Media:
Berlin/Heidelberg, Germany, 2012.
37. Li, Q.; Meng, F.; Yu, Z.; Chen, X.; Huang, Q. Dynamic torso compliance control for standing and walking balance of position-
controlled humanoid robots. IEEE/ASME Trans. Mechatron. 2021, 26, 679–688. [CrossRef]
38. Joint Japanese-French Robotics Laboratory “Eigen-QuadProg”. Available online: https://github.com/jrl-umi3218/eigen-
quadprog (accessed on 12 June 2022).

Micromachines 13 01252

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Micromachines 13 01252

Uploaded by

Copyright:

Available Formats

micromachines

1 School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China

Micromachines 2022, 13, 1252. https://doi.org/10.3390/mi13081252 https://www.mdpi.com/journal/micromachines

2. BHR-W Robot System

Table 1. BHR-W joints angle range.

Joint Angle Range (deg.)

2.2. Hardware System

3. Modeling and Control of Two Locomotion Modes

3.2. Modeling and Control of the Upright Balanced Moving Mode

where ϕ is the model generalized coordinate containing θ, φL and φR .

L( ϕ, ϕ̇) = T ( ϕ, ϕ̇) − V ( ϕ). (2)

where t is the continuous time variable and M is the generalized torque.

Dw ( ϕ) ϕ̈ + Cw( ϕ,ϕ̇) ϕ̇ + Gw (ϕ) =

where ϕ = [θ φL φR ] T , Dw ∈ Rn×n denotes the mass matrix, Cw ∈ Rn denotes the vector

Table 2. Parameters of the upright balanced moving model.

Symbol Parameter Name Value

I{θ,γ} = ∑ mi r{2iy,iz} , (6)

where PCoM ( β) is the position of the CoM, β = [ β 1 β 2 β 3 β 4 ] T is the generalized coordi-

3.2.2. Upright Balanced Moving Control

where x = [θ φL φR θ̇ φ˙L φ˙R ] T , u = [ ML MR ] T , A ∈ R6×6 is the system matrix, B ∈ R6×2 is

Motion planning Upper body motion

ũ = −K̃ x̂. (15)

3.3. Modeling and Control of the Crawling Mode

Table 3. Parameters of the crawling motion model.

Symbol Parameter Name Value

3.3.2. Crawling Motion Control

4. Modeling and Control of Mode Transition

Figure 8. The model of kneeling down.

4.1.2. Kneeling Control

4.2. Modeling and Control of Standing Recovery

Dr (q)q̈ + Cr (q, q̇)q̇ + Gr (q) = Sr τr , (28)

4.2.2. Standing Recovery Control

(a) (b) (c) (d) (e)

From Kneeling to Standing: Based on the simplified underactuated inverted pendulum

A = diag(wq̇1 , wq̇2 , wq̈1 , wq̈2 , wτ1 , wτ2 )

5.1. Upright Balanced Moving Experiment

(a) (b) (c) (d) (e) (f)

5.2. Crawling Experiment

(a) (b) (c) (d)

(e) (f) (g) (h)

5.3. Kneeling Experiment

(a) (b) (c) (d) (e)

5.4. Standing Recovery Experiment

(a) (b) (c) (d) (e)

(a) (b) (c) (d) (e) (f)

(a) (b) (c) (d) (e)

LQR Linear Quadratic Regulator

You might also like

Dw ( ϕ) ϕ̈ + Cw( ϕ,ϕ̇) ϕ̇ + Gw (ϕ) =